This research provides a systematic review of commercial hardware and software accelerators designed specifically for Large Language Models.

• Analyzes the current landscape of proprietary LLM accelerators
• Compares hardware and software characteristics across major commercial accelerators
• Identifies critical design considerations for next-generation accelerators
• Proposes focused research directions for future accelerator development

For engineering teams, this analysis offers valuable insights into optimizing computational infrastructure for AI workloads, potentially reducing costs and improving performance of LLM deployments.

A Review on Proprietary Accelerators for Large Language Models